[FFmpeg-devel] [RFC] An improved implementation?of?ARMv5TE?IDCT (simple_idct_armv5te.S)

Reimar Döffinger Reimar.Doeffinger
Sat Sep 15 10:53:58 CEST 2007


Hello,
On Sat, Sep 15, 2007 at 11:35:43AM +0300, Siarhei Siamashka wrote:
> On 15 September 2007, Michael Niedermayer wrote:
> > On Fri, Sep 14, 2007 at 07:37:01PM +0300, Siarhei Siamashka wrote:
> > [...]
> > > > > +/*
> > > > > + * a local pool with 64-bit constants for 'idct_rows_armv5te'
> > > > > function, + * we align it at 16 byte boundary in order to ensure that
> > > > > it does not cross + * cache line boundary (occupies only a single
> > > > > cache line) + */
> > > > > +        .balign 16
> > > >
> > > > ASMALIGN()
> > >
> > > Does it mean I should define it as a macro at top of
> > > simple_idct_armv5te.S and use it in the rest of the source file?
> >
> > no i mean you should use the ASMALIGN macro as *align is not portable
> > different assemblers interpret it differently or dont support some forms
> > at all
> > i dont know if that is true for assemblers supporting ARM too but ...
> 
> I'm sorry, but please answer my original question. This ASMALIGN() macro
> should be defined somewhere first, otherwise such statement will just result
> in a syntax error. You already rejected a common header file for assembly
> sources, so the only option to define this macro is at the top of
> simple_idct_armv5te.S source. Is that correct? I just wanted to confirm it.

He wouldn't suggest ASMALIGN() without extra explanation if it didn't already exist,
and a quick grep shows it is defined in config.h

Greetings,
Reimar D?ffinger




More information about the ffmpeg-devel mailing list